SUCCESS CASE#MachineLearning #IoT #Edge #AI

Equipment Detection using AI and Edge computing

Cepsa is a consolidated multinational company in the sector, with years of experience and a team of more than 11,000 professionals across the five continents in which it operates, integrating all phases of the energy value chain.

Cepsa aspires to become a benchmark for sustainable mobility, biofuels and green hydrogen in Spain and Portugal, as well as a key company in the Energy Transition, putting customers at the center of its activity and helping them in their efforts towards decarbonization.

AI in Industrial Working Environments.

In industrial environments, the use of safety equipment is necessary to perform certain tasks.

Not using this equipment can increase the number of accidents and the impact of accidents at work. Most current (manual) control systems are expensive and not as effective as desired.

CEPSA is considering replacing manual control mechanisms with solutions based on deep learning models and testing their effectiveness in real environments. Their Data Science team has developed an initial model of image recognition and, in collaboration with Keepler, they are turning this model into a solution with real applicability in their facilities.

The Solution and Main AWS Services Used

Based on the initial deep learning model developed by CEPSA’s data scientists, the challenge for Keepler was to turn this “laboratory” exercise (Data Lab) into a productive, configurable, scalable and easy-to-deploy solution for the various facilities.

A multi-disciplinary Keepler team refined and “productized” it by creating a SaaS (Software as a Service) solution with a serverless cloud architecture and deploying the models on supported devices at the Edge to optimize performance and costs. The solution analyzes the images in real time and sends alerts, via email and SMS, when incidents are detected, in this case, workers without the necessary equipment. Through a very simple web interface, it is possible to add new “locations”, new devices or new users, as well as manage the existing ones.

Keepler adapted the deep learning model to run on different input devices (AI on Edge). The first version can be used on both specific devices (AWS Deeplens) and general purpose ones (Nvidia Jetson Nano), and to ensure its execution in any environment, it was also deployed in a Docker container.

  • The image recognition model was trained and refined using a cloud data science Sandbox.
  • To achieve the best result at the lowest cost, the model was deployed to run locally on compatible devices and was adapted to a camera AWS Deeplens to a module Nvidia Jetson Nano and it was also deployed in a Docker container.
  • SaaS application best practices from the chosen cloud provider were used to ensure that the application took advantage of all the benefits of the public cloud (scalability, security, resilience, reduced cost and operational excellence).

The solution is based entirely on the use of managed services, which provide a serverless implementation that is easy to maintain, robust, secure and scalable. The main Amazon Web Services services used are:

  • AWS DeepLens as a service for the management of the cameras, which has a local model where an AWS Greengrass core is executed.
  • AWS Greengrass for communication between the cloud and edge devices via MQTT, allowing automatic deployment of Machine Learning models.
  • Amazon S3 to store evidence of security incidents (images and metadata), Machine Learning models and the static frontend of the application.
  • AWS Lambda for event processing and REST services that execute the application’s business logic.
  • Amazon API Gateway as a REST API manager (access, throttling, versioning) for communication between frontend and backend services.
  • Amazon Cognito for access control, identity provider and user authentication.
  • Amazon RDS (Aurora Serverless) for storing the solution’s relational information.
  • Amazon CloudWatch for monitoring and operation of infrastructure and log management.
  • AWS IAM for managing access and permissions to the infrastructure.
  • AWS KMS for managing the application’s encryption keys.
Benefits

  • 24/7 system operation.

  • Simple deployment of additional cameras to cover new areas at a reduced incremental cost.

  • The solution allows for the easy deployment of additional deep learning models, making it easier to monitor different equipment or adapt it to other uses.

  • Reduced TCO thanks to the use of AI on the edge and serverless architecture.

  • Incidents are stored, enabling the preparation of statistics for the Security or Risk Prevention departments, and predictive models in the future.

  • The solution is open and allows the connection with other corporate applications to establish correlations between accidents and equipment in order to take corrective measures.

Keepler is a boutique company of professional technology services specialized in design, construction, deployment and software solutions operations of Big Data and Machine Learning for big clients. They use Agile and Devops methodologies and native services of the public cloud to build sophisticated business applications focused in data and integrated with different sources in batch mode and real time. They have Advanced Consulting Partner level and have a technical workforce with 90% of their professionals certified in AWS. Keepler is currently working for big clients in different markets, such as financing services, industry, energy, telecommunications and media.

Let’s talk!

If you want to know more or if you want us to develop a proposal for your specific use, contact us and we’ll talk.